In order to achieve accurate and reliable prediction of the dissolved oxygen in ponds and mitigate aquaculture risks, we propose a predictive model based on Feature Construction （FC） pretreatment and Temporal Convolutional Network （TCN） coupled with Bidirectional Gate Recurrent Unit （BiGRU）. By constructing statistical features, environmental factor features and seasonal features from the samples, deep-level correlations between variables are explored. Then, the structural feature sequences are processed using multiple layers of convolution and dimensionality reduction through TCN, while preserving the global temporal characteristics and removing redundant information. By integrating BiGRU to model the reduced features, accurate prediction of dissolved oxygen levels is achieved. Furthermore, the Sand Cat Swarm Optimization （SCSO） algorithm is employed to optimize the non-parametric Kernel Density Estimation （KDE） for estimating the distribution range of dissolved oxygen prediction errors. The experimental results indicate that the proposed model achieves superior performance compared to other comparative models, with respective values of 0.027 5 for Mean Squared Error （MSE）, 0.143 2 for Mean Absolute Error （MAE）, 0.165 8 for Root Mean Squared Error （RMSE）, and 0.94 for the Coefficient of Determination （R²）. Meanwhile, the interval estimation effectively covers the fluctuation range of dissolved oxygen, thereby quantifying the uncertainty in the prediction process. In the short-term prediction of dissolved oxygen levels in ponds, this model demonstrates notable accuracy and robustness. It is instructive for water quality monitoring in aquaculture and the enhancement of farming efficiency.